Automated electronic document creation through machine learning

Abstract

Aspects of the present disclosure relate to automated electronic document creation. Embodiments include providing, as inputs to a machine learning model, contents of a source electronic document and a prompt. Embodiments include receiving a particular data item from the machine learning model in response to the inputs. Embodiments include performing a lexical search for the particular data item among a plurality of stored data items. Embodiments include determining a confidence score based on the lexical search and performing a semantic search for the particular data item among the plurality of stored data items based on the confidence score not exceeding a threshold. Embodiments include identifying a stored data item of the plurality of data items that matches the particular data item based on the semantic search. Embodiments include automatically generating an electronic document by populating an electronic document template using the stored data item.

What is claimed is: 
     
         1 . A method for automated generation of an electronic document, comprising:
 providing, as inputs to a natural language processing machine learning model, contents of a source electronic document and a prompt that instructs the natural language processing machine learning model to extract a particular data item from the contents of the source electronic document;   receiving the particular data item from the natural language processing machine learning model in response to the inputs;   performing a lexical search for the particular data item among a plurality of stored data items;   determining a confidence score based on the lexical search;   performing a semantic search for the particular data item among the plurality of stored data items based on the confidence score not exceeding a threshold;   identifying a stored data item of the plurality of stored data items that matches the particular data item based on the semantic search; and   automatically generating the electronic document by populating an electronic document template using the stored data item.   
     
     
         2 . The method of  claim 1 , wherein providing the prompt to the natural language processing machine learning model comprises providing a plurality of conditional instructions as contextual information to the natural language processing machine learning model. 
     
     
         3 . The method of  claim 1 , further comprising providing corresponding contents of an associated source electronic document as an additional input to the natural language processing machine learning model, wherein the associated source electronic document is referenced by the source electronic document, and wherein the prompt further instructs the natural language processing machine learning model to extract additional data items from the corresponding contents of the associated source electronic document. 
     
     
         4 . The method of  claim 1 , wherein the performing of the semantic search for the particular data item among the plurality of stored data items based on the confidence score not exceeding the threshold comprises:
 generating an embedding of the particular data item using an embedding model; and   comparing the embedding of the particular data item to corresponding embeddings of the plurality of stored data items.   
     
     
         5 . The method of  claim 1 , further comprising determining a type of the lexical search to perform based on a category associated with the particular data item. 
     
     
         6 . The method of  claim 5 , wherein the determining of the type of the lexical search to perform based on the category of the particular data item comprises determining whether to perform an exact lexical search or a fuzzy lexical search based on a rule related to the category associated with the particular data item. 
     
     
         7 . The method of  claim 1 , further comprising normalizing the confidence score so that the confidence score is between zero and one. 
     
     
         8 . The method of  claim 1 , further comprising applying a boost to the confidence score based on a category associated with the particular data item. 
     
     
         9 . The method of  claim 1 , further comprising:
 receiving feedback with respect to the electronic document via a user interface;   modifying the natural language processing machine learning model or a rule related to the lexical search or the semantic search based on the feedback to produce an updated natural language processing machine learning model or an updated rule; and   automatically generating a subsequent electronic document based on the updated natural language processing machine learning model or the updated rule.   
     
     
         10 . A system for automated creation of an electronic document, comprising:
 one or more processors; and   a memory storing instructions that, when executed by the one or more processors, cause the system to:
 provide, as inputs to a natural language processing machine learning model, contents of a source electronic document and a prompt that instructs the natural language processing machine learning model to extract a particular data item from the contents of the source electronic document; 
 receive the particular data item from the natural language processing machine learning model in response to the inputs; 
 perform a lexical search for the particular data item among a plurality of stored data items; 
 determine a confidence score based on the lexical search; 
 perform a semantic search for the particular data item among the plurality of stored data items based on the confidence score not exceeding a threshold; 
 identify a stored data item of the plurality of stored data items that matches the particular data item based on the semantic search; and 
 automatically generate the electronic document by populating an electronic document template using the stored data item. 
   
     
     
         11 . The system of  claim 10 , wherein providing the prompt to the natural language processing machine learning model comprises providing a plurality of conditional instructions as contextual information to the natural language processing machine learning model. 
     
     
         12 . The system of  claim 10 , wherein the instructions, when executed by the one or more processors, further cause the system to provide corresponding contents of an associated source electronic document as an additional input to the natural language processing machine learning model, wherein the associated source electronic document is referenced by the source electronic document, and wherein the prompt further instructs the natural language processing machine learning model to extract additional data items from the corresponding contents of the associated source electronic document. 
     
     
         13 . The system of  claim 10 , wherein the performing of the semantic search for the particular data item among the plurality of stored data items based on the confidence score not exceeding the threshold comprises:
 generating an embedding of the particular data item using an embedding model; and   comparing the embedding of the particular data item to corresponding embeddings of the plurality of stored data items.   
     
     
         14 . The system of  claim 10 , wherein the instructions, when executed by the one or more processors, further cause the system to determine a type of the lexical search to perform based on a category associated with the particular data item. 
     
     
         15 . The system of  claim 14 , wherein the determining of the type of the lexical search to perform based on the category of the particular data item comprises determining whether to perform an exact lexical search or a fuzzy lexical search based on a rule related to the category associated with the particular data item. 
     
     
         16 . The system of  claim 10 , wherein the instructions, when executed by the one or more processors, further cause the system to normalize the confidence score so that the confidence score is between zero and one. 
     
     
         17 . The system of  claim 10  wherein the instructions, when executed by the one or more processors, further cause the system to apply a boost to the confidence score based on a category associated with the particular data item. 
     
     
         18 . The system of  claim 10 , wherein the instructions, when executed by the one or more processors, further cause the system to:
 receive feedback with respect to the electronic document via a user interface;   modify the natural language processing machine learning model or a rule related to the lexical search or the semantic search based on the feedback to produce an updated natural language processing machine learning model or an updated rule; and   automatically generate a subsequent electronic document based on the updated natural language processing machine learning model or the updated rule.   
     
     
         19 . A non-transitory computer readable medium comprising instructions that, when executed by one or more processors of a computing system, cause the computing system to:
 provide, as inputs to a natural language processing machine learning model, contents of a source electronic document and a prompt that instructs the natural language processing machine learning model to extract a particular data item from the contents of the source electronic document;   receive the particular data item from the natural language processing machine learning model in response to the inputs;   perform a lexical search for the particular data item among a plurality of stored data items;   determine a confidence score based on the lexical search;   perform a semantic search for the particular data item among the plurality of stored data items based on the confidence score not exceeding a threshold;   identify a stored data item of the plurality of stored data items that matches the particular data item based on the semantic search; and   automatically generate the electronic document by populating an electronic document template using the stored data item.   
     
     
         20 . The non-transitory computer readable medium of  claim 19 , wherein providing the prompt to the natural language processing machine learning model comprises providing a plurality of conditional instructions as contextual information to the natural language processing machine learning model.